I noticed that the number of trees in the fftb forrest for some of my test items (for inferred grammars with lots of ambiguity) doesn’t necessarily match the number of parses reported by the parse file when I processed a profile with ACE. The difference is 1908 from ACE vs 2448 in fftb. Should these numbers be the same? If not, what’s the difference?
The numbers are not expected to match. They are both kind of supposed to measure the same thing, but the number given by FFTB is an estimate produced without actually enumerating the readings and seeing which ones are valid. It corresponds to the number of ways of unpacking the forest to form trees consistent with what was found during the packing phase of parsing. Some of those can prove to be inconsistent (i.e. imply failed unifications) upon closer inspection. This happens, for instance, when a more specific edge packs into a more general edge, which then undergoes a rule which is compatible with the more general edge but not the more specific one. If that sounds like gobbledygook I can try to find a more concrete example…
Thanks, Woodley. I don’t understand perfectly, but well enough for my purposes